Path Planning of Unmanned Ships Based on Deep Reinforcement Learning
XU Wenkai
SUI Jianghua
YANG Shukui
LYU Han
PENG Yunjie
Abstract：For the unmanned ship path planning problem in complex sea environment,a path planning algorithm based on the priority experience replay dueling deep Q network(PER Dueling DQN)is proposed,which transforms the action value function of traditional deep Q network(DQN)into the state value function as well as the action dominance function.In addition,a priority expe-rience replay mechanism is introduced to prioritize TD-error's as experience samples to enhance the sampling rate of key experience data.The experimental results show that PER Dueling DQN improves the convergence speed by 50%and shortens the total path length by 16.5%in the random grid environment compared with the traditional DQN.The convergence speed improves by 42.85%and shortens the total path length by 1.73%in the sea area environment.The study shows that PER Dueling DQN is effectively appli-cable to a variety of sea area environments,which provides the unmanned ship with the ability to achieve autonomous path planning tasks in complex sea areas.The study shows that PER Dueling DQN is effectively applied to a variety of sea environments,which provides some theoretical support for the autonomous path planning task of unmanned vessels in complex sea areas.
Keywords：unmanned shipspath planningdeep Q networkpriority experience playbackcompetitive network structure
Publication Date：2025-10-20
Online Publishing Date：2026-01-16(First online date of this platform, not the publication date of the document)
Pages：6( 49-54 )